Dear Gina, 

is this not merely a statistical problem?

http://oak.ucc.nau.edu/rh232/courses/EPS525/Handouts/Understanding%20the%20Independent%20t%20Test.pdf

The rigorous ways to analyze standards vs. deviants are (1) downsample your standards until the same number of trials are included as of deviants, and (2) compare responses to the same stimulus when it's serving as the standard and as the deviant.

Downsampling until you have a matched number of trials is a good sanity-check any time you have differing numbers of trials/blocks/whatever per condition. If your effect is still there, it's probably 'real'.

I note that this issue and issues related to counterbalancing can also arise when for whatever reason runs are rejected due to artifact.

I would like to add that for most of the fMRI data analysis approaches in the field, one builds a regressor (i.e. expected time series) by convolving the event onset w.r.t. a canonical HRF. Strictly speaking, a regressor corresponding to a single trial presentation contains the same expected data point as compared to one corresponding to the presentation of 10/20/30 trials. If the single trial presentation is eliciting reliable and robust BOLD activations, from the regression analysis you'll get a beta as good as those regressors with multiple trials presentation. So, yes, the number of trials between different conditions will definitely influence the statistical power of your analysis, but it might not be so detrimental as you imagine, especially if a single trial event is robust and reliablely activating a specific region of the brain. An Odd-ball task is one good example.

Thanks to everyone for responding. I just wanted to summarize the answers I have received from consulting different researchers, including statisticians, in hopes that this will be of use for others out there who are not in memory research (where different number of events per condition is inherent to the paradigms used, as analyses depend on participant responses, and this issue as an issue never comes up).

Having an uneven number of events between conditions should not influence what effects you have found, that is, it does not invalidate any of the statistical tests and results. It can only compromise the magnitude of the contrasts you test, or statistical power -- which is not an issue if the effects being tested are strong. Although comparing conditions with different number of trials is not statistical invalid (unless variances differ, then it's a violation of an assumption of parametric tests, and not directly due to uneven number of events), one could use the limiting/lowest number of events in any given condition and randomly select events from the other conditions to match (as is explained in more detail by a couple of posts above). From what I understand, this is superfluous but serves as a check just to check or just to keep things "clean".